Re: Should we add crc32 in libpgport?

On Thu, Feb 16, 2012 at 6:09 AM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Fri, Feb 3, 2012 at 7:33 PM, Daniel Farina <daniel(at)heroku(dot)com> wrote:
>> Ah, yes, I think my optimizations were off when building, or
>> something.  I didn't get such verbosity at first, and then I remember
>> doing something slightly different and then getting a lot of output.
>> I didn't pay attention to the build size.  I will investigate.
> [...]
>>
>> I agree, I was about to say "what about a preprocessor hack..." after
>> the last paragraph, but saw you beat me to the punch.  I'll have a look soon.
>
> Ping!

Err, yes. Clearly I've managed to not do this, and not see your email
until now. Here's what I intend to do:

1) Split the tables into another header file, per Tom's suggestion

2) #include those tables in pgport exactly once. Per Tom's objection
that pgport is not always available in distributions, that is not the
only way the table will be exposed, but as pgport is definitely built
and available when building postgres proper.

3) Third-party projects and contribs should use the header file, and
not libpgport

It's still a bit awkward in that one is including something that's not
really a "port" (except in the degenerate sense, as no system has
these tables defined vs, say, gettimeofday, where Windows needs a
port), but it's the only thing that I can see that is compiled once
and can be linked against repeatedly in postgres's build without
having to, say, directly cross-reference the pg_crc.o file (as seen in
the two command line utilities that need crc).

Thoughts?

--
fdr